Rearch And Application On SVM-based Obstacle Recognition For High-voltage Power Line Inspection Robot

There is realistic meaning to ensure the safe operation of the circuit and improve the efficiency of power line maintenance in using inspection robot replacing traditional way of power transmission line inspection and maintenance. The obstacle recognition is one of critical technologies for inspection robot to achieve completely intelligent. The inspection robot must be able to identify the spacer, counter weight and other line hardware in order to move automatically on the power line and finish a wider range of inspection work. For 500 KV quaternary fission power line, the paper designs an obstacle recognition algorithm based on SVM for the power transmission intelligent inspection robot, which can assist inspection robot to indentify some typical hardware on the line.This paper has carried on the research analysis for high voltage transmission lines, and analyses the spacer, counter weight and other hardware features and installation characteristics..According to the inspection robot obstacle-navigation method, it selects four kinds of hardware: spacer and shock hammer, insulator and suspension clamp as identifying object and put forward an SVM–based obstacle recognition algorithm. The algorithm consists of two parts: feature extraction module and classification module.It carried on noise reduction processing by median filtering method for collected obstacles images before the feature extraction. Feature extraction is an important process of target recognition. This paper selects HOG features of target image as image description of obstacles, gives HOG feature extraction process, and extracts HOG features of four kinds of typical hardware and gives its image expression. In order to accelerate the algorithm running speed, it conducts PCA analysis for the extraction of the HOG characteristics. It puts forward using sparse matrix to carry on dimension reduction for HOG feature, which is based on PCA dimension reduction.In this paper, it presents SVM–based multi–classification obstacle recognition algorithm. Based on the principles of linear classification, it introduces support vector machine and multiple classification. First, it uses cross validation method to obtain the optimal SVM parameters, carries on comparative analysis for several common SVM–based multi–classification methods, and puts forward using the decision tree method to design obstacle recognition system. Then, it presents code–based decision classification method, which is based on one-against-one classification. Through the contrast experiment, it gets four kinds of hardware best samples combination scheme under different classification strategy. It makes the characteristics as variable carry on comparison experiment, and analyzes the effect of HOG characteristics and SPS- HOG features to the classification result. Finally, it respectively uses decision tree method and coding system carried on obstacle recognition under the optimal parameters. The experimental results show that the SVM–based obstacle recognition algorithm proposed in this paper has high identification accuracy and good real-time performance, which can meet the requirements of inspection robot obstacle recognition.